AI is becoming Alaska's front page. It only knows the Alaska that's online.

The first AI answer many Alaskans see today does not come from opening ChatGPT. It shows up inside an ordinary Google search, above the links, before anyone has clicked a source.

Type a name, a district, or an issue into Google, and an AI-written summary may sit at the top of the page. Most people do not think of that as using artificial intelligence. They think they are searching. But that box comes from the same kind of model that powers a chatbot. Google says these AI Overviews now reach 1.5 billion people a month, across 200 countries and territories.

This is the quiet shift. AI is not replacing search. It is becoming the search result. And it does not act like a neutral directory. It works more like a compression layer over the web. It summarizes what is already written, what ranks high, what is recent, and what has enough online coverage to show up at all. In Alaska, that last part is the catch.

Two modes: summarize, or go quiet

A benchmark test ran this from inside Alaska. It asked the kind of question any resident deserves a clear answer to: who is running, and where do they stand.

From a clean browser in Valdez, Google's AI summaries showed up for the higher-profile statewide questions. They hedged, listed candidates, and leaned toward names with recent, well-covered records. But ask about the local state senate district, from inside that district, and no AI summary appeared at all. The page fell back to plain links, several of them about other races. Ask directly who to vote for, and it returned nothing.

That is the whole pattern in miniature. Where the coverage exists, the machine summarizes. Where it does not, the machine goes quiet.

The silence is measurable, and it tracks the news deserts

That test runs every day, across six consumer AI models, on voter-style questions about Alaska's 2026 races. The pattern holds. For the U.S. Senate race at the top of the ballot, the models refused to name any candidate 18 percent of the time. For the governor's race and a state senate district, they stayed silent more than half the time.

The silence is not the machine being careful. It is the machine reading a map. High-profile Alaska races have candidate pages, national mentions, and fresh links. Down-ballot races, in a state that has lost much of its local press, often have almost nothing for a model to read. The AI knows what the web knows. And the web does not cover Alaska evenly.

What the answer is really made of

Here is the part that should give pause to anyone who reads an AI summary as a verdict.

The answer is mostly not the model's judgment. It is retrieval. In one test, a single model changed which candidate it named on a contested cost-of-living question based only on how many web searches it was allowed to run. A shallow search surfaced one candidate. A deeper search surfaced another. The sample is small, one contested race, so hold it loosely. But the direction points to something bigger.